2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
6 * This handles basic git sha1 object files - packing, unpacking,
10 #include "string-list.h"
15 #include "run-command.h"
18 #include "tree-walk.h"
20 #include "pack-revindex.h"
21 #include "sha1-lookup.h"
22 #include "bulk-checkin.h"
23 #include "streaming.h"
27 #if defined(__linux__) && (defined(__i386__) || defined(__PPC__))
28 #define O_NOATIME 01000000
34 #define SZ_FMT PRIuMAX
35 static inline uintmax_t sz_fmt(size_t s
) { return s
; }
37 const unsigned char null_sha1
[20];
39 static const char *no_log_pack_access
= "no_log_pack_access";
40 static const char *log_pack_access
;
43 * This is meant to hold a *small* number of objects that you would
44 * want read_sha1_file() to be able to return, but yet you do not want
45 * to write them into the object store (e.g. a browse-only
48 static struct cached_object
{
49 unsigned char sha1
[20];
50 enum object_type type
;
54 static int cached_object_nr
, cached_object_alloc
;
56 static struct cached_object empty_tree
= {
57 EMPTY_TREE_SHA1_BIN_LITERAL
,
63 static struct packed_git
*last_found_pack
;
65 static struct cached_object
*find_cached_object(const unsigned char *sha1
)
68 struct cached_object
*co
= cached_objects
;
70 for (i
= 0; i
< cached_object_nr
; i
++, co
++) {
71 if (!hashcmp(co
->sha1
, sha1
))
74 if (!hashcmp(sha1
, empty_tree
.sha1
))
79 int mkdir_in_gitdir(const char *path
)
81 if (mkdir(path
, 0777)) {
82 int saved_errno
= errno
;
84 struct strbuf sb
= STRBUF_INIT
;
89 * Are we looking at a path in a symlinked worktree
90 * whose original repository does not yet have it?
91 * e.g. .git/rr-cache pointing at its original
92 * repository in which the user hasn't performed any
93 * conflict resolution yet?
95 if (lstat(path
, &st
) || !S_ISLNK(st
.st_mode
) ||
96 strbuf_readlink(&sb
, path
, st
.st_size
) ||
97 !is_absolute_path(sb
.buf
) ||
98 mkdir(sb
.buf
, 0777)) {
105 return adjust_shared_perm(path
);
108 int safe_create_leading_directories(char *path
)
110 char *pos
= path
+ offset_1st_component(path
);
114 pos
= strchr(pos
, '/');
117 while (*++pos
== '/')
122 if (!stat(path
, &st
)) {
124 if (!S_ISDIR(st
.st_mode
)) {
129 else if (mkdir(path
, 0777)) {
130 if (errno
== EEXIST
&&
131 !stat(path
, &st
) && S_ISDIR(st
.st_mode
)) {
132 ; /* somebody created it since we checked */
138 else if (adjust_shared_perm(path
)) {
147 int safe_create_leading_directories_const(const char *path
)
149 /* path points to cache entries, so xstrdup before messing with it */
150 char *buf
= xstrdup(path
);
151 int result
= safe_create_leading_directories(buf
);
156 static void fill_sha1_path(char *pathbuf
, const unsigned char *sha1
)
159 for (i
= 0; i
< 20; i
++) {
160 static char hex
[] = "0123456789abcdef";
161 unsigned int val
= sha1
[i
];
162 char *pos
= pathbuf
+ i
*2 + (i
> 0);
163 *pos
++ = hex
[val
>> 4];
164 *pos
= hex
[val
& 0xf];
169 * NOTE! This returns a statically allocated buffer, so you have to be
170 * careful about using it. Do an "xstrdup()" if you need to save the
173 * Also note that this returns the location for creating. Reading
174 * SHA1 file can happen from any alternate directory listed in the
175 * DB_ENVIRONMENT environment variable if it is not found in
176 * the primary object database.
178 char *sha1_file_name(const unsigned char *sha1
)
180 static char buf
[PATH_MAX
];
184 objdir
= get_object_directory();
185 len
= strlen(objdir
);
187 /* '/' + sha1(2) + '/' + sha1(38) + '\0' */
188 if (len
+ 43 > PATH_MAX
)
189 die("insanely long object directory %s", objdir
);
190 memcpy(buf
, objdir
, len
);
194 fill_sha1_path(buf
+ len
+ 1, sha1
);
198 static char *sha1_get_pack_name(const unsigned char *sha1
,
199 char **name
, char **base
, const char *which
)
201 static const char hex
[] = "0123456789abcdef";
206 const char *sha1_file_directory
= get_object_directory();
207 int len
= strlen(sha1_file_directory
);
208 *base
= xmalloc(len
+ 60);
209 sprintf(*base
, "%s/pack/pack-1234567890123456789012345678901234567890.%s",
210 sha1_file_directory
, which
);
211 *name
= *base
+ len
+ 11;
216 for (i
= 0; i
< 20; i
++) {
217 unsigned int val
= *sha1
++;
218 *buf
++ = hex
[val
>> 4];
219 *buf
++ = hex
[val
& 0xf];
225 char *sha1_pack_name(const unsigned char *sha1
)
227 static char *name
, *base
;
229 return sha1_get_pack_name(sha1
, &name
, &base
, "pack");
232 char *sha1_pack_index_name(const unsigned char *sha1
)
234 static char *name
, *base
;
236 return sha1_get_pack_name(sha1
, &name
, &base
, "idx");
239 struct alternate_object_database
*alt_odb_list
;
240 static struct alternate_object_database
**alt_odb_tail
;
242 static int git_open_noatime(const char *name
);
245 * Prepare alternate object database registry.
247 * The variable alt_odb_list points at the list of struct
248 * alternate_object_database. The elements on this list come from
249 * non-empty elements from colon separated ALTERNATE_DB_ENVIRONMENT
250 * environment variable, and $GIT_OBJECT_DIRECTORY/info/alternates,
251 * whose contents is similar to that environment variable but can be
252 * LF separated. Its base points at a statically allocated buffer that
253 * contains "/the/directory/corresponding/to/.git/objects/...", while
254 * its name points just after the slash at the end of ".git/objects/"
255 * in the example above, and has enough space to hold 40-byte hex
256 * SHA1, an extra slash for the first level indirection, and the
259 static int link_alt_odb_entry(const char *entry
, const char *relative_base
, int depth
)
261 const char *objdir
= get_object_directory();
262 struct alternate_object_database
*ent
;
263 struct alternate_object_database
*alt
;
265 struct strbuf pathbuf
= STRBUF_INIT
;
267 if (!is_absolute_path(entry
) && relative_base
) {
268 strbuf_addstr(&pathbuf
, real_path(relative_base
));
269 strbuf_addch(&pathbuf
, '/');
271 strbuf_addstr(&pathbuf
, entry
);
273 normalize_path_copy(pathbuf
.buf
, pathbuf
.buf
);
275 pfxlen
= strlen(pathbuf
.buf
);
278 * The trailing slash after the directory name is given by
279 * this function at the end. Remove duplicates.
281 while (pfxlen
&& pathbuf
.buf
[pfxlen
-1] == '/')
284 entlen
= pfxlen
+ 43; /* '/' + 2 hex + '/' + 38 hex + NUL */
285 ent
= xmalloc(sizeof(*ent
) + entlen
);
286 memcpy(ent
->base
, pathbuf
.buf
, pfxlen
);
287 strbuf_release(&pathbuf
);
289 ent
->name
= ent
->base
+ pfxlen
+ 1;
290 ent
->base
[pfxlen
+ 3] = '/';
291 ent
->base
[pfxlen
] = ent
->base
[entlen
-1] = 0;
293 /* Detect cases where alternate disappeared */
294 if (!is_directory(ent
->base
)) {
295 error("object directory %s does not exist; "
296 "check .git/objects/info/alternates.",
302 /* Prevent the common mistake of listing the same
303 * thing twice, or object directory itself.
305 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
306 if (!memcmp(ent
->base
, alt
->base
, pfxlen
)) {
311 if (!strcmp(ent
->base
, objdir
)) {
316 /* add the alternate entry */
318 alt_odb_tail
= &(ent
->next
);
321 /* recursively add alternates */
322 read_info_alternates(ent
->base
, depth
+ 1);
324 ent
->base
[pfxlen
] = '/';
329 static void link_alt_odb_entries(const char *alt
, int len
, int sep
,
330 const char *relative_base
, int depth
)
332 struct string_list entries
= STRING_LIST_INIT_NODUP
;
337 error("%s: ignoring alternate object stores, nesting too deep.",
342 alt_copy
= xmemdupz(alt
, len
);
343 string_list_split_in_place(&entries
, alt_copy
, sep
, -1);
344 for (i
= 0; i
< entries
.nr
; i
++) {
345 const char *entry
= entries
.items
[i
].string
;
346 if (entry
[0] == '\0' || entry
[0] == '#')
348 if (!is_absolute_path(entry
) && depth
) {
349 error("%s: ignoring relative alternate object store %s",
350 relative_base
, entry
);
352 link_alt_odb_entry(entry
, relative_base
, depth
);
355 string_list_clear(&entries
, 0);
359 void read_info_alternates(const char * relative_base
, int depth
)
364 const char alt_file_name
[] = "info/alternates";
365 /* Given that relative_base is no longer than PATH_MAX,
366 ensure that "path" has enough space to append "/", the
367 file name, "info/alternates", and a trailing NUL. */
368 char path
[PATH_MAX
+ 1 + sizeof alt_file_name
];
371 sprintf(path
, "%s/%s", relative_base
, alt_file_name
);
372 fd
= git_open_noatime(path
);
375 if (fstat(fd
, &st
) || (st
.st_size
== 0)) {
379 mapsz
= xsize_t(st
.st_size
);
380 map
= xmmap(NULL
, mapsz
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
383 link_alt_odb_entries(map
, mapsz
, '\n', relative_base
, depth
);
388 void add_to_alternates_file(const char *reference
)
390 struct lock_file
*lock
= xcalloc(1, sizeof(struct lock_file
));
391 int fd
= hold_lock_file_for_append(lock
, git_path("objects/info/alternates"), LOCK_DIE_ON_ERROR
);
392 char *alt
= mkpath("%s\n", reference
);
393 write_or_die(fd
, alt
, strlen(alt
));
394 if (commit_lock_file(lock
))
395 die("could not close alternates file");
397 link_alt_odb_entries(alt
, strlen(alt
), '\n', NULL
, 0);
400 void foreach_alt_odb(alt_odb_fn fn
, void *cb
)
402 struct alternate_object_database
*ent
;
405 for (ent
= alt_odb_list
; ent
; ent
= ent
->next
)
410 void prepare_alt_odb(void)
417 alt
= getenv(ALTERNATE_DB_ENVIRONMENT
);
420 alt_odb_tail
= &alt_odb_list
;
421 link_alt_odb_entries(alt
, strlen(alt
), PATH_SEP
, NULL
, 0);
423 read_info_alternates(get_object_directory(), 0);
426 static int has_loose_object_local(const unsigned char *sha1
)
428 char *name
= sha1_file_name(sha1
);
429 return !access(name
, F_OK
);
432 int has_loose_object_nonlocal(const unsigned char *sha1
)
434 struct alternate_object_database
*alt
;
436 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
437 fill_sha1_path(alt
->name
, sha1
);
438 if (!access(alt
->base
, F_OK
))
444 static int has_loose_object(const unsigned char *sha1
)
446 return has_loose_object_local(sha1
) ||
447 has_loose_object_nonlocal(sha1
);
450 static unsigned int pack_used_ctr
;
451 static unsigned int pack_mmap_calls
;
452 static unsigned int peak_pack_open_windows
;
453 static unsigned int pack_open_windows
;
454 static unsigned int pack_open_fds
;
455 static unsigned int pack_max_fds
;
456 static size_t peak_pack_mapped
;
457 static size_t pack_mapped
;
458 struct packed_git
*packed_git
;
460 void pack_report(void)
463 "pack_report: getpagesize() = %10" SZ_FMT
"\n"
464 "pack_report: core.packedGitWindowSize = %10" SZ_FMT
"\n"
465 "pack_report: core.packedGitLimit = %10" SZ_FMT
"\n",
466 sz_fmt(getpagesize()),
467 sz_fmt(packed_git_window_size
),
468 sz_fmt(packed_git_limit
));
470 "pack_report: pack_used_ctr = %10u\n"
471 "pack_report: pack_mmap_calls = %10u\n"
472 "pack_report: pack_open_windows = %10u / %10u\n"
473 "pack_report: pack_mapped = "
474 "%10" SZ_FMT
" / %10" SZ_FMT
"\n",
477 pack_open_windows
, peak_pack_open_windows
,
478 sz_fmt(pack_mapped
), sz_fmt(peak_pack_mapped
));
481 static int check_packed_git_idx(const char *path
, struct packed_git
*p
)
484 struct pack_idx_header
*hdr
;
486 uint32_t version
, nr
, i
, *index
;
487 int fd
= git_open_noatime(path
);
492 if (fstat(fd
, &st
)) {
496 idx_size
= xsize_t(st
.st_size
);
497 if (idx_size
< 4 * 256 + 20 + 20) {
499 return error("index file %s is too small", path
);
501 idx_map
= xmmap(NULL
, idx_size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
505 if (hdr
->idx_signature
== htonl(PACK_IDX_SIGNATURE
)) {
506 version
= ntohl(hdr
->idx_version
);
507 if (version
< 2 || version
> 2) {
508 munmap(idx_map
, idx_size
);
509 return error("index file %s is version %"PRIu32
510 " and is not supported by this binary"
511 " (try upgrading GIT to a newer version)",
520 index
+= 2; /* skip index header */
521 for (i
= 0; i
< 256; i
++) {
522 uint32_t n
= ntohl(index
[i
]);
524 munmap(idx_map
, idx_size
);
525 return error("non-monotonic index %s", path
);
533 * - 256 index entries 4 bytes each
534 * - 24-byte entries * nr (20-byte sha1 + 4-byte offset)
535 * - 20-byte SHA1 of the packfile
536 * - 20-byte SHA1 file checksum
538 if (idx_size
!= 4*256 + nr
* 24 + 20 + 20) {
539 munmap(idx_map
, idx_size
);
540 return error("wrong index v1 file size in %s", path
);
542 } else if (version
== 2) {
545 * - 8 bytes of header
546 * - 256 index entries 4 bytes each
547 * - 20-byte sha1 entry * nr
548 * - 4-byte crc entry * nr
549 * - 4-byte offset entry * nr
550 * - 20-byte SHA1 of the packfile
551 * - 20-byte SHA1 file checksum
552 * And after the 4-byte offset table might be a
553 * variable sized table containing 8-byte entries
554 * for offsets larger than 2^31.
556 unsigned long min_size
= 8 + 4*256 + nr
*(20 + 4 + 4) + 20 + 20;
557 unsigned long max_size
= min_size
;
559 max_size
+= (nr
- 1)*8;
560 if (idx_size
< min_size
|| idx_size
> max_size
) {
561 munmap(idx_map
, idx_size
);
562 return error("wrong index v2 file size in %s", path
);
564 if (idx_size
!= min_size
&&
566 * make sure we can deal with large pack offsets.
567 * 31-bit signed offset won't be enough, neither
568 * 32-bit unsigned one will be.
570 (sizeof(off_t
) <= 4)) {
571 munmap(idx_map
, idx_size
);
572 return error("pack too large for current definition of off_t in %s", path
);
576 p
->index_version
= version
;
577 p
->index_data
= idx_map
;
578 p
->index_size
= idx_size
;
583 int open_pack_index(struct packed_git
*p
)
591 idx_name
= xstrdup(p
->pack_name
);
592 strcpy(idx_name
+ strlen(idx_name
) - strlen(".pack"), ".idx");
593 ret
= check_packed_git_idx(idx_name
, p
);
598 static void scan_windows(struct packed_git
*p
,
599 struct packed_git
**lru_p
,
600 struct pack_window
**lru_w
,
601 struct pack_window
**lru_l
)
603 struct pack_window
*w
, *w_l
;
605 for (w_l
= NULL
, w
= p
->windows
; w
; w
= w
->next
) {
607 if (!*lru_w
|| w
->last_used
< (*lru_w
)->last_used
) {
617 static int unuse_one_window(struct packed_git
*current
)
619 struct packed_git
*p
, *lru_p
= NULL
;
620 struct pack_window
*lru_w
= NULL
, *lru_l
= NULL
;
623 scan_windows(current
, &lru_p
, &lru_w
, &lru_l
);
624 for (p
= packed_git
; p
; p
= p
->next
)
625 scan_windows(p
, &lru_p
, &lru_w
, &lru_l
);
627 munmap(lru_w
->base
, lru_w
->len
);
628 pack_mapped
-= lru_w
->len
;
630 lru_l
->next
= lru_w
->next
;
632 lru_p
->windows
= lru_w
->next
;
640 void release_pack_memory(size_t need
)
642 size_t cur
= pack_mapped
;
643 while (need
>= (cur
- pack_mapped
) && unuse_one_window(NULL
))
647 void *xmmap(void *start
, size_t length
,
648 int prot
, int flags
, int fd
, off_t offset
)
650 void *ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
651 if (ret
== MAP_FAILED
) {
654 release_pack_memory(length
);
655 ret
= mmap(start
, length
, prot
, flags
, fd
, offset
);
656 if (ret
== MAP_FAILED
)
657 die_errno("Out of memory? mmap failed");
662 void close_pack_windows(struct packed_git
*p
)
665 struct pack_window
*w
= p
->windows
;
668 die("pack '%s' still has open windows to it",
670 munmap(w
->base
, w
->len
);
671 pack_mapped
-= w
->len
;
673 p
->windows
= w
->next
;
679 * The LRU pack is the one with the oldest MRU window, preferring packs
680 * with no used windows, or the oldest mtime if it has no windows allocated.
682 static void find_lru_pack(struct packed_git
*p
, struct packed_git
**lru_p
, struct pack_window
**mru_w
, int *accept_windows_inuse
)
684 struct pack_window
*w
, *this_mru_w
;
685 int has_windows_inuse
= 0;
688 * Reject this pack if it has windows and the previously selected
689 * one does not. If this pack does not have windows, reject
690 * it if the pack file is newer than the previously selected one.
692 if (*lru_p
&& !*mru_w
&& (p
->windows
|| p
->mtime
> (*lru_p
)->mtime
))
695 for (w
= this_mru_w
= p
->windows
; w
; w
= w
->next
) {
697 * Reject this pack if any of its windows are in use,
698 * but the previously selected pack did not have any
699 * inuse windows. Otherwise, record that this pack
700 * has windows in use.
703 if (*accept_windows_inuse
)
704 has_windows_inuse
= 1;
709 if (w
->last_used
> this_mru_w
->last_used
)
713 * Reject this pack if it has windows that have been
714 * used more recently than the previously selected pack.
715 * If the previously selected pack had windows inuse and
716 * we have not encountered a window in this pack that is
717 * inuse, skip this check since we prefer a pack with no
718 * inuse windows to one that has inuse windows.
720 if (*mru_w
&& *accept_windows_inuse
== has_windows_inuse
&&
721 this_mru_w
->last_used
> (*mru_w
)->last_used
)
730 *accept_windows_inuse
= has_windows_inuse
;
733 static int close_one_pack(void)
735 struct packed_git
*p
, *lru_p
= NULL
;
736 struct pack_window
*mru_w
= NULL
;
737 int accept_windows_inuse
= 1;
739 for (p
= packed_git
; p
; p
= p
->next
) {
740 if (p
->pack_fd
== -1)
742 find_lru_pack(p
, &lru_p
, &mru_w
, &accept_windows_inuse
);
746 close(lru_p
->pack_fd
);
755 void unuse_pack(struct pack_window
**w_cursor
)
757 struct pack_window
*w
= *w_cursor
;
764 void close_pack_index(struct packed_git
*p
)
767 munmap((void *)p
->index_data
, p
->index_size
);
768 p
->index_data
= NULL
;
773 * This is used by git-repack in case a newly created pack happens to
774 * contain the same set of objects as an existing one. In that case
775 * the resulting file might be different even if its name would be the
776 * same. It is best to close any reference to the old pack before it is
777 * replaced on disk. Of course no index pointers nor windows for given pack
778 * must subsist at this point. If ever objects from this pack are requested
779 * again, the new version of the pack will be reinitialized through
780 * reprepare_packed_git().
782 void free_pack_by_name(const char *pack_name
)
784 struct packed_git
*p
, **pp
= &packed_git
;
788 if (strcmp(pack_name
, p
->pack_name
) == 0) {
789 clear_delta_base_cache();
790 close_pack_windows(p
);
791 if (p
->pack_fd
!= -1) {
796 free(p
->bad_object_sha1
);
798 if (last_found_pack
== p
)
799 last_found_pack
= NULL
;
807 static unsigned int get_max_fd_limit(void)
812 if (getrlimit(RLIMIT_NOFILE
, &lim
))
813 die_errno("cannot get RLIMIT_NOFILE");
816 #elif defined(_SC_OPEN_MAX)
817 return sysconf(_SC_OPEN_MAX
);
818 #elif defined(OPEN_MAX)
821 return 1; /* see the caller ;-) */
826 * Do not call this directly as this leaks p->pack_fd on error return;
827 * call open_packed_git() instead.
829 static int open_packed_git_1(struct packed_git
*p
)
832 struct pack_header hdr
;
833 unsigned char sha1
[20];
834 unsigned char *idx_sha1
;
837 if (!p
->index_data
&& open_pack_index(p
))
838 return error("packfile %s index unavailable", p
->pack_name
);
841 unsigned int max_fds
= get_max_fd_limit();
843 /* Save 3 for stdin/stdout/stderr, 22 for work */
845 pack_max_fds
= max_fds
- 25;
850 while (pack_max_fds
<= pack_open_fds
&& close_one_pack())
853 p
->pack_fd
= git_open_noatime(p
->pack_name
);
854 if (p
->pack_fd
< 0 || fstat(p
->pack_fd
, &st
))
858 /* If we created the struct before we had the pack we lack size. */
860 if (!S_ISREG(st
.st_mode
))
861 return error("packfile %s not a regular file", p
->pack_name
);
862 p
->pack_size
= st
.st_size
;
863 } else if (p
->pack_size
!= st
.st_size
)
864 return error("packfile %s size changed", p
->pack_name
);
866 /* We leave these file descriptors open with sliding mmap;
867 * there is no point keeping them open across exec(), though.
869 fd_flag
= fcntl(p
->pack_fd
, F_GETFD
, 0);
871 return error("cannot determine file descriptor flags");
872 fd_flag
|= FD_CLOEXEC
;
873 if (fcntl(p
->pack_fd
, F_SETFD
, fd_flag
) == -1)
874 return error("cannot set FD_CLOEXEC");
876 /* Verify we recognize this pack file format. */
877 if (read_in_full(p
->pack_fd
, &hdr
, sizeof(hdr
)) != sizeof(hdr
))
878 return error("file %s is far too short to be a packfile", p
->pack_name
);
879 if (hdr
.hdr_signature
!= htonl(PACK_SIGNATURE
))
880 return error("file %s is not a GIT packfile", p
->pack_name
);
881 if (!pack_version_ok(hdr
.hdr_version
))
882 return error("packfile %s is version %"PRIu32
" and not"
883 " supported (try upgrading GIT to a newer version)",
884 p
->pack_name
, ntohl(hdr
.hdr_version
));
886 /* Verify the pack matches its index. */
887 if (p
->num_objects
!= ntohl(hdr
.hdr_entries
))
888 return error("packfile %s claims to have %"PRIu32
" objects"
889 " while index indicates %"PRIu32
" objects",
890 p
->pack_name
, ntohl(hdr
.hdr_entries
),
892 if (lseek(p
->pack_fd
, p
->pack_size
- sizeof(sha1
), SEEK_SET
) == -1)
893 return error("end of packfile %s is unavailable", p
->pack_name
);
894 if (read_in_full(p
->pack_fd
, sha1
, sizeof(sha1
)) != sizeof(sha1
))
895 return error("packfile %s signature is unavailable", p
->pack_name
);
896 idx_sha1
= ((unsigned char *)p
->index_data
) + p
->index_size
- 40;
897 if (hashcmp(sha1
, idx_sha1
))
898 return error("packfile %s does not match index", p
->pack_name
);
902 static int open_packed_git(struct packed_git
*p
)
904 if (!open_packed_git_1(p
))
906 if (p
->pack_fd
!= -1) {
914 static int in_window(struct pack_window
*win
, off_t offset
)
916 /* We must promise at least 20 bytes (one hash) after the
917 * offset is available from this window, otherwise the offset
918 * is not actually in this window and a different window (which
919 * has that one hash excess) must be used. This is to support
920 * the object header and delta base parsing routines below.
922 off_t win_off
= win
->offset
;
923 return win_off
<= offset
924 && (offset
+ 20) <= (win_off
+ win
->len
);
927 unsigned char *use_pack(struct packed_git
*p
,
928 struct pack_window
**w_cursor
,
932 struct pack_window
*win
= *w_cursor
;
934 /* Since packfiles end in a hash of their content and it's
935 * pointless to ask for an offset into the middle of that
936 * hash, and the in_window function above wouldn't match
937 * don't allow an offset too close to the end of the file.
939 if (!p
->pack_size
&& p
->pack_fd
== -1 && open_packed_git(p
))
940 die("packfile %s cannot be accessed", p
->pack_name
);
941 if (offset
> (p
->pack_size
- 20))
942 die("offset beyond end of packfile (truncated pack?)");
944 if (!win
|| !in_window(win
, offset
)) {
947 for (win
= p
->windows
; win
; win
= win
->next
) {
948 if (in_window(win
, offset
))
952 size_t window_align
= packed_git_window_size
/ 2;
955 if (p
->pack_fd
== -1 && open_packed_git(p
))
956 die("packfile %s cannot be accessed", p
->pack_name
);
958 win
= xcalloc(1, sizeof(*win
));
959 win
->offset
= (offset
/ window_align
) * window_align
;
960 len
= p
->pack_size
- win
->offset
;
961 if (len
> packed_git_window_size
)
962 len
= packed_git_window_size
;
963 win
->len
= (size_t)len
;
964 pack_mapped
+= win
->len
;
965 while (packed_git_limit
< pack_mapped
966 && unuse_one_window(p
))
968 win
->base
= xmmap(NULL
, win
->len
,
969 PROT_READ
, MAP_PRIVATE
,
970 p
->pack_fd
, win
->offset
);
971 if (win
->base
== MAP_FAILED
)
972 die("packfile %s cannot be mapped: %s",
975 if (!win
->offset
&& win
->len
== p
->pack_size
976 && !p
->do_not_close
) {
983 if (pack_mapped
> peak_pack_mapped
)
984 peak_pack_mapped
= pack_mapped
;
985 if (pack_open_windows
> peak_pack_open_windows
)
986 peak_pack_open_windows
= pack_open_windows
;
987 win
->next
= p
->windows
;
991 if (win
!= *w_cursor
) {
992 win
->last_used
= pack_used_ctr
++;
996 offset
-= win
->offset
;
998 *left
= win
->len
- xsize_t(offset
);
999 return win
->base
+ offset
;
1002 static struct packed_git
*alloc_packed_git(int extra
)
1004 struct packed_git
*p
= xmalloc(sizeof(*p
) + extra
);
1005 memset(p
, 0, sizeof(*p
));
1010 static void try_to_free_pack_memory(size_t size
)
1012 release_pack_memory(size
);
1015 struct packed_git
*add_packed_git(const char *path
, int path_len
, int local
)
1017 static int have_set_try_to_free_routine
;
1019 struct packed_git
*p
= alloc_packed_git(path_len
+ 2);
1021 if (!have_set_try_to_free_routine
) {
1022 have_set_try_to_free_routine
= 1;
1023 set_try_to_free_routine(try_to_free_pack_memory
);
1027 * Make sure a corresponding .pack file exists and that
1028 * the index looks sane.
1030 path_len
-= strlen(".idx");
1035 memcpy(p
->pack_name
, path
, path_len
);
1037 strcpy(p
->pack_name
+ path_len
, ".keep");
1038 if (!access(p
->pack_name
, F_OK
))
1041 strcpy(p
->pack_name
+ path_len
, ".pack");
1042 if (stat(p
->pack_name
, &st
) || !S_ISREG(st
.st_mode
)) {
1047 /* ok, it looks sane as far as we can check without
1048 * actually mapping the pack file.
1050 p
->pack_size
= st
.st_size
;
1051 p
->pack_local
= local
;
1052 p
->mtime
= st
.st_mtime
;
1053 if (path_len
< 40 || get_sha1_hex(path
+ path_len
- 40, p
->sha1
))
1058 struct packed_git
*parse_pack_index(unsigned char *sha1
, const char *idx_path
)
1060 const char *path
= sha1_pack_name(sha1
);
1061 struct packed_git
*p
= alloc_packed_git(strlen(path
) + 1);
1063 strcpy(p
->pack_name
, path
);
1064 hashcpy(p
->sha1
, sha1
);
1065 if (check_packed_git_idx(idx_path
, p
)) {
1073 void install_packed_git(struct packed_git
*pack
)
1075 if (pack
->pack_fd
!= -1)
1078 pack
->next
= packed_git
;
1082 void (*report_garbage
)(const char *desc
, const char *path
);
1084 static void report_helper(const struct string_list
*list
,
1085 int seen_bits
, int first
, int last
)
1088 switch (seen_bits
) {
1090 msg
= "no corresponding .idx nor .pack";
1093 msg
= "no corresponding .idx";
1096 msg
= "no corresponding .pack";
1101 for (; first
< last
; first
++)
1102 report_garbage(msg
, list
->items
[first
].string
);
1105 static void report_pack_garbage(struct string_list
*list
)
1107 int i
, baselen
= -1, first
= 0, seen_bits
= 0;
1109 if (!report_garbage
)
1112 sort_string_list(list
);
1114 for (i
= 0; i
< list
->nr
; i
++) {
1115 const char *path
= list
->items
[i
].string
;
1116 if (baselen
!= -1 &&
1117 strncmp(path
, list
->items
[first
].string
, baselen
)) {
1118 report_helper(list
, seen_bits
, first
, i
);
1122 if (baselen
== -1) {
1123 const char *dot
= strrchr(path
, '.');
1125 report_garbage("garbage found", path
);
1128 baselen
= dot
- path
+ 1;
1131 if (!strcmp(path
+ baselen
, "pack"))
1133 else if (!strcmp(path
+ baselen
, "idx"))
1136 report_helper(list
, seen_bits
, first
, list
->nr
);
1139 static void prepare_packed_git_one(char *objdir
, int local
)
1141 /* Ensure that this buffer is large enough so that we can
1142 append "/pack/" without clobbering the stack even if
1143 strlen(objdir) were PATH_MAX. */
1144 char path
[PATH_MAX
+ 1 + 4 + 1 + 1];
1148 struct string_list garbage
= STRING_LIST_INIT_DUP
;
1150 sprintf(path
, "%s/pack", objdir
);
1152 dir
= opendir(path
);
1154 if (errno
!= ENOENT
)
1155 error("unable to open object pack directory: %s: %s",
1156 path
, strerror(errno
));
1160 while ((de
= readdir(dir
)) != NULL
) {
1161 int namelen
= strlen(de
->d_name
);
1162 struct packed_git
*p
;
1164 if (len
+ namelen
+ 1 > sizeof(path
)) {
1165 if (report_garbage
) {
1166 struct strbuf sb
= STRBUF_INIT
;
1167 strbuf_addf(&sb
, "%.*s/%s", len
- 1, path
, de
->d_name
);
1168 report_garbage("path too long", sb
.buf
);
1169 strbuf_release(&sb
);
1174 if (is_dot_or_dotdot(de
->d_name
))
1177 strcpy(path
+ len
, de
->d_name
);
1179 if (has_extension(de
->d_name
, ".idx")) {
1180 /* Don't reopen a pack we already have. */
1181 for (p
= packed_git
; p
; p
= p
->next
) {
1182 if (!memcmp(path
, p
->pack_name
, len
+ namelen
- 4))
1187 * See if it really is a valid .idx file with
1188 * corresponding .pack file that we can map.
1190 (p
= add_packed_git(path
, len
+ namelen
, local
)) != NULL
)
1191 install_packed_git(p
);
1194 if (!report_garbage
)
1197 if (has_extension(de
->d_name
, ".idx") ||
1198 has_extension(de
->d_name
, ".pack") ||
1199 has_extension(de
->d_name
, ".keep"))
1200 string_list_append(&garbage
, path
);
1202 report_garbage("garbage found", path
);
1205 report_pack_garbage(&garbage
);
1206 string_list_clear(&garbage
, 0);
1209 static int sort_pack(const void *a_
, const void *b_
)
1211 struct packed_git
*a
= *((struct packed_git
**)a_
);
1212 struct packed_git
*b
= *((struct packed_git
**)b_
);
1216 * Local packs tend to contain objects specific to our
1217 * variant of the project than remote ones. In addition,
1218 * remote ones could be on a network mounted filesystem.
1219 * Favor local ones for these reasons.
1221 st
= a
->pack_local
- b
->pack_local
;
1226 * Younger packs tend to contain more recent objects,
1227 * and more recent objects tend to get accessed more
1230 if (a
->mtime
< b
->mtime
)
1232 else if (a
->mtime
== b
->mtime
)
1237 static void rearrange_packed_git(void)
1239 struct packed_git
**ary
, *p
;
1242 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1247 /* prepare an array of packed_git for easier sorting */
1248 ary
= xcalloc(n
, sizeof(struct packed_git
*));
1249 for (n
= 0, p
= packed_git
; p
; p
= p
->next
)
1252 qsort(ary
, n
, sizeof(struct packed_git
*), sort_pack
);
1254 /* link them back again */
1255 for (i
= 0; i
< n
- 1; i
++)
1256 ary
[i
]->next
= ary
[i
+ 1];
1257 ary
[n
- 1]->next
= NULL
;
1258 packed_git
= ary
[0];
1263 static int prepare_packed_git_run_once
= 0;
1264 void prepare_packed_git(void)
1266 struct alternate_object_database
*alt
;
1268 if (prepare_packed_git_run_once
)
1270 prepare_packed_git_one(get_object_directory(), 1);
1272 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1274 prepare_packed_git_one(alt
->base
, 0);
1275 alt
->name
[-1] = '/';
1277 rearrange_packed_git();
1278 prepare_packed_git_run_once
= 1;
1281 void reprepare_packed_git(void)
1284 prepare_packed_git_run_once
= 0;
1285 prepare_packed_git();
1288 static void mark_bad_packed_object(struct packed_git
*p
,
1289 const unsigned char *sha1
)
1292 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1293 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1295 p
->bad_object_sha1
= xrealloc(p
->bad_object_sha1
, 20 * (p
->num_bad_objects
+ 1));
1296 hashcpy(p
->bad_object_sha1
+ 20 * p
->num_bad_objects
, sha1
);
1297 p
->num_bad_objects
++;
1300 static const struct packed_git
*has_packed_and_bad(const unsigned char *sha1
)
1302 struct packed_git
*p
;
1305 for (p
= packed_git
; p
; p
= p
->next
)
1306 for (i
= 0; i
< p
->num_bad_objects
; i
++)
1307 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
1313 * With an in-core object data in "map", rehash it to make sure the
1314 * object name actually matches "sha1" to detect object corruption.
1315 * With "map" == NULL, try reading the object named with "sha1" using
1316 * the streaming interface and rehash it to do the same.
1318 int check_sha1_signature(const unsigned char *sha1
, void *map
,
1319 unsigned long size
, const char *type
)
1321 unsigned char real_sha1
[20];
1322 enum object_type obj_type
;
1323 struct git_istream
*st
;
1329 hash_sha1_file(map
, size
, type
, real_sha1
);
1330 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1333 st
= open_istream(sha1
, &obj_type
, &size
, NULL
);
1337 /* Generate the header */
1338 hdrlen
= sprintf(hdr
, "%s %lu", typename(obj_type
), size
) + 1;
1342 git_SHA1_Update(&c
, hdr
, hdrlen
);
1344 char buf
[1024 * 16];
1345 ssize_t readlen
= read_istream(st
, buf
, sizeof(buf
));
1353 git_SHA1_Update(&c
, buf
, readlen
);
1355 git_SHA1_Final(real_sha1
, &c
);
1357 return hashcmp(sha1
, real_sha1
) ? -1 : 0;
1360 static int git_open_noatime(const char *name
)
1362 static int sha1_file_open_flag
= O_NOATIME
;
1365 int fd
= open(name
, O_RDONLY
| sha1_file_open_flag
);
1369 /* Might the failure be due to O_NOATIME? */
1370 if (errno
!= ENOENT
&& sha1_file_open_flag
) {
1371 sha1_file_open_flag
= 0;
1379 static int stat_sha1_file(const unsigned char *sha1
, struct stat
*st
)
1381 char *name
= sha1_file_name(sha1
);
1382 struct alternate_object_database
*alt
;
1384 if (!lstat(name
, st
))
1389 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1391 fill_sha1_path(name
, sha1
);
1392 if (!lstat(alt
->base
, st
))
1399 static int open_sha1_file(const unsigned char *sha1
)
1402 char *name
= sha1_file_name(sha1
);
1403 struct alternate_object_database
*alt
;
1405 fd
= git_open_noatime(name
);
1411 for (alt
= alt_odb_list
; alt
; alt
= alt
->next
) {
1413 fill_sha1_path(name
, sha1
);
1414 fd
= git_open_noatime(alt
->base
);
1421 void *map_sha1_file(const unsigned char *sha1
, unsigned long *size
)
1426 fd
= open_sha1_file(sha1
);
1431 if (!fstat(fd
, &st
)) {
1432 *size
= xsize_t(st
.st_size
);
1434 /* mmap() is forbidden on empty files */
1435 error("object file %s is empty", sha1_file_name(sha1
));
1438 map
= xmmap(NULL
, *size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
1446 * There used to be a second loose object header format which
1447 * was meant to mimic the in-pack format, allowing for direct
1448 * copy of the object data. This format turned up not to be
1449 * really worth it and we no longer write loose objects in that
1452 static int experimental_loose_object(unsigned char *map
)
1457 * We must determine if the buffer contains the standard
1458 * zlib-deflated stream or the experimental format based
1459 * on the in-pack object format. Compare the header byte
1462 * RFC1950 zlib w/ deflate : 0www1000 : 0 <= www <= 7
1463 * Experimental pack-based : Stttssss : ttt = 1,2,3,4
1465 * If bit 7 is clear and bits 0-3 equal 8, the buffer MUST be
1466 * in standard loose-object format, UNLESS it is a Git-pack
1467 * format object *exactly* 8 bytes in size when inflated.
1469 * However, RFC1950 also specifies that the 1st 16-bit word
1470 * must be divisible by 31 - this checksum tells us our buffer
1471 * is in the standard format, giving a false positive only if
1472 * the 1st word of the Git-pack format object happens to be
1473 * divisible by 31, ie:
1474 * ((byte0 * 256) + byte1) % 31 = 0
1475 * => 0ttt10000www1000 % 31 = 0
1477 * As it happens, this case can only arise for www=3 & ttt=1
1478 * - ie, a Commit object, which would have to be 8 bytes in
1479 * size. As no Commit can be that small, we find that the
1480 * combination of these two criteria (bitmask & checksum)
1481 * can always correctly determine the buffer format.
1483 word
= (map
[0] << 8) + map
[1];
1484 if ((map
[0] & 0x8F) == 0x08 && !(word
% 31))
1490 unsigned long unpack_object_header_buffer(const unsigned char *buf
,
1491 unsigned long len
, enum object_type
*type
, unsigned long *sizep
)
1494 unsigned long size
, c
;
1495 unsigned long used
= 0;
1498 *type
= (c
>> 4) & 7;
1502 if (len
<= used
|| bitsizeof(long) <= shift
) {
1503 error("bad object header");
1508 size
+= (c
& 0x7f) << shift
;
1515 int unpack_sha1_header(git_zstream
*stream
, unsigned char *map
, unsigned long mapsize
, void *buffer
, unsigned long bufsiz
)
1517 unsigned long size
, used
;
1518 static const char valid_loose_object_type
[8] = {
1520 1, 1, 1, 1, /* "commit", "tree", "blob", "tag" */
1521 0, /* "delta" and others are invalid in a loose object */
1523 enum object_type type
;
1525 /* Get the data stream */
1526 memset(stream
, 0, sizeof(*stream
));
1527 stream
->next_in
= map
;
1528 stream
->avail_in
= mapsize
;
1529 stream
->next_out
= buffer
;
1530 stream
->avail_out
= bufsiz
;
1532 if (experimental_loose_object(map
)) {
1534 * The old experimental format we no longer produce;
1535 * we can still read it.
1537 used
= unpack_object_header_buffer(map
, mapsize
, &type
, &size
);
1538 if (!used
|| !valid_loose_object_type
[type
])
1543 /* Set up the stream for the rest.. */
1544 stream
->next_in
= map
;
1545 stream
->avail_in
= mapsize
;
1546 git_inflate_init(stream
);
1548 /* And generate the fake traditional header */
1549 stream
->total_out
= 1 + snprintf(buffer
, bufsiz
, "%s %lu",
1550 typename(type
), size
);
1553 git_inflate_init(stream
);
1554 return git_inflate(stream
, 0);
1557 static void *unpack_sha1_rest(git_zstream
*stream
, void *buffer
, unsigned long size
, const unsigned char *sha1
)
1559 int bytes
= strlen(buffer
) + 1;
1560 unsigned char *buf
= xmallocz(size
);
1564 n
= stream
->total_out
- bytes
;
1567 memcpy(buf
, (char *) buffer
+ bytes
, n
);
1569 if (bytes
<= size
) {
1571 * The above condition must be (bytes <= size), not
1572 * (bytes < size). In other words, even though we
1573 * expect no more output and set avail_out to zero,
1574 * the input zlib stream may have bytes that express
1575 * "this concludes the stream", and we *do* want to
1578 * Otherwise we would not be able to test that we
1579 * consumed all the input to reach the expected size;
1580 * we also want to check that zlib tells us that all
1581 * went well with status == Z_STREAM_END at the end.
1583 stream
->next_out
= buf
+ bytes
;
1584 stream
->avail_out
= size
- bytes
;
1585 while (status
== Z_OK
)
1586 status
= git_inflate(stream
, Z_FINISH
);
1588 if (status
== Z_STREAM_END
&& !stream
->avail_in
) {
1589 git_inflate_end(stream
);
1594 error("corrupt loose object '%s'", sha1_to_hex(sha1
));
1595 else if (stream
->avail_in
)
1596 error("garbage at end of loose object '%s'",
1603 * We used to just use "sscanf()", but that's actually way
1604 * too permissive for what we want to check. So do an anal
1605 * object header parse by hand.
1607 int parse_sha1_header(const char *hdr
, unsigned long *sizep
)
1614 * The type can be at most ten bytes (including the
1615 * terminating '\0' that we add), and is followed by
1624 if (i
>= sizeof(type
))
1630 * The length must follow immediately, and be in canonical
1631 * decimal format (ie "010" is not valid).
1633 size
= *hdr
++ - '0';
1638 unsigned long c
= *hdr
- '0';
1642 size
= size
* 10 + c
;
1648 * The length must be followed by a zero byte
1650 return *hdr
? -1 : type_from_string(type
);
1653 static void *unpack_sha1_file(void *map
, unsigned long mapsize
, enum object_type
*type
, unsigned long *size
, const unsigned char *sha1
)
1659 ret
= unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
));
1660 if (ret
< Z_OK
|| (*type
= parse_sha1_header(hdr
, size
)) < 0)
1663 return unpack_sha1_rest(&stream
, hdr
, *size
, sha1
);
1666 unsigned long get_size_from_delta(struct packed_git
*p
,
1667 struct pack_window
**w_curs
,
1670 const unsigned char *data
;
1671 unsigned char delta_head
[20], *in
;
1675 memset(&stream
, 0, sizeof(stream
));
1676 stream
.next_out
= delta_head
;
1677 stream
.avail_out
= sizeof(delta_head
);
1679 git_inflate_init(&stream
);
1681 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1682 stream
.next_in
= in
;
1683 st
= git_inflate(&stream
, Z_FINISH
);
1684 curpos
+= stream
.next_in
- in
;
1685 } while ((st
== Z_OK
|| st
== Z_BUF_ERROR
) &&
1686 stream
.total_out
< sizeof(delta_head
));
1687 git_inflate_end(&stream
);
1688 if ((st
!= Z_STREAM_END
) && stream
.total_out
!= sizeof(delta_head
)) {
1689 error("delta data unpack-initial failed");
1693 /* Examine the initial part of the delta to figure out
1698 /* ignore base size */
1699 get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1701 /* Read the result size */
1702 return get_delta_hdr_size(&data
, delta_head
+sizeof(delta_head
));
1705 static off_t
get_delta_base(struct packed_git
*p
,
1706 struct pack_window
**w_curs
,
1708 enum object_type type
,
1709 off_t delta_obj_offset
)
1711 unsigned char *base_info
= use_pack(p
, w_curs
, *curpos
, NULL
);
1714 /* use_pack() assured us we have [base_info, base_info + 20)
1715 * as a range that we can look at without walking off the
1716 * end of the mapped window. Its actually the hash size
1717 * that is assured. An OFS_DELTA longer than the hash size
1718 * is stupid, as then a REF_DELTA would be smaller to store.
1720 if (type
== OBJ_OFS_DELTA
) {
1722 unsigned char c
= base_info
[used
++];
1723 base_offset
= c
& 127;
1726 if (!base_offset
|| MSB(base_offset
, 7))
1727 return 0; /* overflow */
1728 c
= base_info
[used
++];
1729 base_offset
= (base_offset
<< 7) + (c
& 127);
1731 base_offset
= delta_obj_offset
- base_offset
;
1732 if (base_offset
<= 0 || base_offset
>= delta_obj_offset
)
1733 return 0; /* out of bound */
1735 } else if (type
== OBJ_REF_DELTA
) {
1736 /* The base entry _must_ be in the same pack */
1737 base_offset
= find_pack_entry_one(base_info
, p
);
1740 die("I am totally screwed");
1744 int unpack_object_header(struct packed_git
*p
,
1745 struct pack_window
**w_curs
,
1747 unsigned long *sizep
)
1749 unsigned char *base
;
1752 enum object_type type
;
1754 /* use_pack() assures us we have [base, base + 20) available
1755 * as a range that we can look at. (Its actually the hash
1756 * size that is assured.) With our object header encoding
1757 * the maximum deflated object size is 2^137, which is just
1758 * insane, so we know won't exceed what we have been given.
1760 base
= use_pack(p
, w_curs
, *curpos
, &left
);
1761 used
= unpack_object_header_buffer(base
, left
, &type
, sizep
);
1770 static int retry_bad_packed_offset(struct packed_git
*p
, off_t obj_offset
)
1773 struct revindex_entry
*revidx
;
1774 const unsigned char *sha1
;
1775 revidx
= find_pack_revindex(p
, obj_offset
);
1778 sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
1779 mark_bad_packed_object(p
, sha1
);
1780 type
= sha1_object_info(sha1
, NULL
);
1781 if (type
<= OBJ_NONE
)
1786 #define POI_STACK_PREALLOC 64
1788 static enum object_type
packed_to_object_type(struct packed_git
*p
,
1790 enum object_type type
,
1791 struct pack_window
**w_curs
,
1794 off_t small_poi_stack
[POI_STACK_PREALLOC
];
1795 off_t
*poi_stack
= small_poi_stack
;
1796 int poi_stack_nr
= 0, poi_stack_alloc
= POI_STACK_PREALLOC
;
1798 while (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1801 /* Push the object we're going to leave behind */
1802 if (poi_stack_nr
>= poi_stack_alloc
&& poi_stack
== small_poi_stack
) {
1803 poi_stack_alloc
= alloc_nr(poi_stack_nr
);
1804 poi_stack
= xmalloc(sizeof(off_t
)*poi_stack_alloc
);
1805 memcpy(poi_stack
, small_poi_stack
, sizeof(off_t
)*poi_stack_nr
);
1807 ALLOC_GROW(poi_stack
, poi_stack_nr
+1, poi_stack_alloc
);
1809 poi_stack
[poi_stack_nr
++] = obj_offset
;
1810 /* If parsing the base offset fails, just unwind */
1811 base_offset
= get_delta_base(p
, w_curs
, &curpos
, type
, obj_offset
);
1814 curpos
= obj_offset
= base_offset
;
1815 type
= unpack_object_header(p
, w_curs
, &curpos
, &size
);
1816 if (type
<= OBJ_NONE
) {
1817 /* If getting the base itself fails, we first
1818 * retry the base, otherwise unwind */
1819 type
= retry_bad_packed_offset(p
, base_offset
);
1820 if (type
> OBJ_NONE
)
1834 error("unknown object type %i at offset %"PRIuMAX
" in %s",
1835 type
, (uintmax_t)obj_offset
, p
->pack_name
);
1840 if (poi_stack
!= small_poi_stack
)
1845 while (poi_stack_nr
) {
1846 obj_offset
= poi_stack
[--poi_stack_nr
];
1847 type
= retry_bad_packed_offset(p
, obj_offset
);
1848 if (type
> OBJ_NONE
)
1855 static int packed_object_info(struct packed_git
*p
, off_t obj_offset
,
1856 struct object_info
*oi
)
1858 struct pack_window
*w_curs
= NULL
;
1860 off_t curpos
= obj_offset
;
1861 enum object_type type
;
1864 * We always get the representation type, but only convert it to
1865 * a "real" type later if the caller is interested.
1867 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
1870 if (type
== OBJ_OFS_DELTA
|| type
== OBJ_REF_DELTA
) {
1871 off_t tmp_pos
= curpos
;
1872 off_t base_offset
= get_delta_base(p
, &w_curs
, &tmp_pos
,
1878 *oi
->sizep
= get_size_from_delta(p
, &w_curs
, tmp_pos
);
1879 if (*oi
->sizep
== 0) {
1888 if (oi
->disk_sizep
) {
1889 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
1890 *oi
->disk_sizep
= revidx
[1].offset
- obj_offset
;
1894 *oi
->typep
= packed_to_object_type(p
, obj_offset
, type
, &w_curs
, curpos
);
1895 if (*oi
->typep
< 0) {
1902 unuse_pack(&w_curs
);
1906 static void *unpack_compressed_entry(struct packed_git
*p
,
1907 struct pack_window
**w_curs
,
1913 unsigned char *buffer
, *in
;
1915 buffer
= xmallocz(size
);
1916 memset(&stream
, 0, sizeof(stream
));
1917 stream
.next_out
= buffer
;
1918 stream
.avail_out
= size
+ 1;
1920 git_inflate_init(&stream
);
1922 in
= use_pack(p
, w_curs
, curpos
, &stream
.avail_in
);
1923 stream
.next_in
= in
;
1924 st
= git_inflate(&stream
, Z_FINISH
);
1925 if (!stream
.avail_out
)
1926 break; /* the payload is larger than it should be */
1927 curpos
+= stream
.next_in
- in
;
1928 } while (st
== Z_OK
|| st
== Z_BUF_ERROR
);
1929 git_inflate_end(&stream
);
1930 if ((st
!= Z_STREAM_END
) || stream
.total_out
!= size
) {
1938 #define MAX_DELTA_CACHE (256)
1940 static size_t delta_base_cached
;
1942 static struct delta_base_cache_lru_list
{
1943 struct delta_base_cache_lru_list
*prev
;
1944 struct delta_base_cache_lru_list
*next
;
1945 } delta_base_cache_lru
= { &delta_base_cache_lru
, &delta_base_cache_lru
};
1947 static struct delta_base_cache_entry
{
1948 struct delta_base_cache_lru_list lru
;
1950 struct packed_git
*p
;
1953 enum object_type type
;
1954 } delta_base_cache
[MAX_DELTA_CACHE
];
1956 static unsigned long pack_entry_hash(struct packed_git
*p
, off_t base_offset
)
1960 hash
= (unsigned long)p
+ (unsigned long)base_offset
;
1961 hash
+= (hash
>> 8) + (hash
>> 16);
1962 return hash
% MAX_DELTA_CACHE
;
1965 static struct delta_base_cache_entry
*
1966 get_delta_base_cache_entry(struct packed_git
*p
, off_t base_offset
)
1968 unsigned long hash
= pack_entry_hash(p
, base_offset
);
1969 return delta_base_cache
+ hash
;
1972 static int eq_delta_base_cache_entry(struct delta_base_cache_entry
*ent
,
1973 struct packed_git
*p
, off_t base_offset
)
1975 return (ent
->data
&& ent
->p
== p
&& ent
->base_offset
== base_offset
);
1978 static int in_delta_base_cache(struct packed_git
*p
, off_t base_offset
)
1980 struct delta_base_cache_entry
*ent
;
1981 ent
= get_delta_base_cache_entry(p
, base_offset
);
1982 return eq_delta_base_cache_entry(ent
, p
, base_offset
);
1985 static void clear_delta_base_cache_entry(struct delta_base_cache_entry
*ent
)
1988 ent
->lru
.next
->prev
= ent
->lru
.prev
;
1989 ent
->lru
.prev
->next
= ent
->lru
.next
;
1990 delta_base_cached
-= ent
->size
;
1993 static void *cache_or_unpack_entry(struct packed_git
*p
, off_t base_offset
,
1994 unsigned long *base_size
, enum object_type
*type
, int keep_cache
)
1996 struct delta_base_cache_entry
*ent
;
1999 ent
= get_delta_base_cache_entry(p
, base_offset
);
2001 if (!eq_delta_base_cache_entry(ent
, p
, base_offset
))
2002 return unpack_entry(p
, base_offset
, type
, base_size
);
2007 clear_delta_base_cache_entry(ent
);
2009 ret
= xmemdupz(ent
->data
, ent
->size
);
2011 *base_size
= ent
->size
;
2015 static inline void release_delta_base_cache(struct delta_base_cache_entry
*ent
)
2020 ent
->lru
.next
->prev
= ent
->lru
.prev
;
2021 ent
->lru
.prev
->next
= ent
->lru
.next
;
2022 delta_base_cached
-= ent
->size
;
2026 void clear_delta_base_cache(void)
2029 for (p
= 0; p
< MAX_DELTA_CACHE
; p
++)
2030 release_delta_base_cache(&delta_base_cache
[p
]);
2033 static void add_delta_base_cache(struct packed_git
*p
, off_t base_offset
,
2034 void *base
, unsigned long base_size
, enum object_type type
)
2036 unsigned long hash
= pack_entry_hash(p
, base_offset
);
2037 struct delta_base_cache_entry
*ent
= delta_base_cache
+ hash
;
2038 struct delta_base_cache_lru_list
*lru
;
2040 release_delta_base_cache(ent
);
2041 delta_base_cached
+= base_size
;
2043 for (lru
= delta_base_cache_lru
.next
;
2044 delta_base_cached
> delta_base_cache_limit
2045 && lru
!= &delta_base_cache_lru
;
2047 struct delta_base_cache_entry
*f
= (void *)lru
;
2048 if (f
->type
== OBJ_BLOB
)
2049 release_delta_base_cache(f
);
2051 for (lru
= delta_base_cache_lru
.next
;
2052 delta_base_cached
> delta_base_cache_limit
2053 && lru
!= &delta_base_cache_lru
;
2055 struct delta_base_cache_entry
*f
= (void *)lru
;
2056 release_delta_base_cache(f
);
2060 ent
->base_offset
= base_offset
;
2063 ent
->size
= base_size
;
2064 ent
->lru
.next
= &delta_base_cache_lru
;
2065 ent
->lru
.prev
= delta_base_cache_lru
.prev
;
2066 delta_base_cache_lru
.prev
->next
= &ent
->lru
;
2067 delta_base_cache_lru
.prev
= &ent
->lru
;
2070 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2071 unsigned long *size
);
2073 static void write_pack_access_log(struct packed_git
*p
, off_t obj_offset
)
2075 static FILE *log_file
;
2077 if (!log_pack_access
)
2078 log_pack_access
= getenv("GIT_TRACE_PACK_ACCESS");
2079 if (!log_pack_access
)
2080 log_pack_access
= no_log_pack_access
;
2081 if (log_pack_access
== no_log_pack_access
)
2085 log_file
= fopen(log_pack_access
, "w");
2087 error("cannot open pack access log '%s' for writing: %s",
2088 log_pack_access
, strerror(errno
));
2089 log_pack_access
= no_log_pack_access
;
2093 fprintf(log_file
, "%s %"PRIuMAX
"\n",
2094 p
->pack_name
, (uintmax_t)obj_offset
);
2098 int do_check_packed_object_crc
;
2100 #define UNPACK_ENTRY_STACK_PREALLOC 64
2101 struct unpack_entry_stack_ent
{
2107 void *unpack_entry(struct packed_git
*p
, off_t obj_offset
,
2108 enum object_type
*final_type
, unsigned long *final_size
)
2110 struct pack_window
*w_curs
= NULL
;
2111 off_t curpos
= obj_offset
;
2114 enum object_type type
;
2115 struct unpack_entry_stack_ent small_delta_stack
[UNPACK_ENTRY_STACK_PREALLOC
];
2116 struct unpack_entry_stack_ent
*delta_stack
= small_delta_stack
;
2117 int delta_stack_nr
= 0, delta_stack_alloc
= UNPACK_ENTRY_STACK_PREALLOC
;
2118 int base_from_cache
= 0;
2120 if (log_pack_access
!= no_log_pack_access
)
2121 write_pack_access_log(p
, obj_offset
);
2123 /* PHASE 1: drill down to the innermost base object */
2127 struct delta_base_cache_entry
*ent
;
2129 ent
= get_delta_base_cache_entry(p
, curpos
);
2130 if (eq_delta_base_cache_entry(ent
, p
, curpos
)) {
2134 clear_delta_base_cache_entry(ent
);
2135 base_from_cache
= 1;
2139 if (do_check_packed_object_crc
&& p
->index_version
> 1) {
2140 struct revindex_entry
*revidx
= find_pack_revindex(p
, obj_offset
);
2141 unsigned long len
= revidx
[1].offset
- obj_offset
;
2142 if (check_pack_crc(p
, &w_curs
, obj_offset
, len
, revidx
->nr
)) {
2143 const unsigned char *sha1
=
2144 nth_packed_object_sha1(p
, revidx
->nr
);
2145 error("bad packed object CRC for %s",
2147 mark_bad_packed_object(p
, sha1
);
2148 unuse_pack(&w_curs
);
2153 type
= unpack_object_header(p
, &w_curs
, &curpos
, &size
);
2154 if (type
!= OBJ_OFS_DELTA
&& type
!= OBJ_REF_DELTA
)
2157 base_offset
= get_delta_base(p
, &w_curs
, &curpos
, type
, obj_offset
);
2159 error("failed to validate delta base reference "
2160 "at offset %"PRIuMAX
" from %s",
2161 (uintmax_t)curpos
, p
->pack_name
);
2162 /* bail to phase 2, in hopes of recovery */
2167 /* push object, proceed to base */
2168 if (delta_stack_nr
>= delta_stack_alloc
2169 && delta_stack
== small_delta_stack
) {
2170 delta_stack_alloc
= alloc_nr(delta_stack_nr
);
2171 delta_stack
= xmalloc(sizeof(*delta_stack
)*delta_stack_alloc
);
2172 memcpy(delta_stack
, small_delta_stack
,
2173 sizeof(*delta_stack
)*delta_stack_nr
);
2175 ALLOC_GROW(delta_stack
, delta_stack_nr
+1, delta_stack_alloc
);
2177 i
= delta_stack_nr
++;
2178 delta_stack
[i
].obj_offset
= obj_offset
;
2179 delta_stack
[i
].curpos
= curpos
;
2180 delta_stack
[i
].size
= size
;
2182 curpos
= obj_offset
= base_offset
;
2185 /* PHASE 2: handle the base */
2190 die("BUG in unpack_entry: left loop at a valid delta");
2196 if (!base_from_cache
)
2197 data
= unpack_compressed_entry(p
, &w_curs
, curpos
, size
);
2201 error("unknown object type %i at offset %"PRIuMAX
" in %s",
2202 type
, (uintmax_t)obj_offset
, p
->pack_name
);
2205 /* PHASE 3: apply deltas in order */
2208 * 'data' holds the base data, or NULL if there was corruption
2210 while (delta_stack_nr
) {
2213 unsigned long delta_size
, base_size
= size
;
2219 add_delta_base_cache(p
, obj_offset
, base
, base_size
, type
);
2223 * We're probably in deep shit, but let's try to fetch
2224 * the required base anyway from another pack or loose.
2225 * This is costly but should happen only in the presence
2226 * of a corrupted pack, and is better than failing outright.
2228 struct revindex_entry
*revidx
;
2229 const unsigned char *base_sha1
;
2230 revidx
= find_pack_revindex(p
, obj_offset
);
2232 base_sha1
= nth_packed_object_sha1(p
, revidx
->nr
);
2233 error("failed to read delta base object %s"
2234 " at offset %"PRIuMAX
" from %s",
2235 sha1_to_hex(base_sha1
), (uintmax_t)obj_offset
,
2237 mark_bad_packed_object(p
, base_sha1
);
2238 base
= read_object(base_sha1
, &type
, &base_size
);
2242 i
= --delta_stack_nr
;
2243 obj_offset
= delta_stack
[i
].obj_offset
;
2244 curpos
= delta_stack
[i
].curpos
;
2245 delta_size
= delta_stack
[i
].size
;
2250 delta_data
= unpack_compressed_entry(p
, &w_curs
, curpos
, delta_size
);
2253 error("failed to unpack compressed delta "
2254 "at offset %"PRIuMAX
" from %s",
2255 (uintmax_t)curpos
, p
->pack_name
);
2260 data
= patch_delta(base
, base_size
,
2261 delta_data
, delta_size
,
2265 * We could not apply the delta; warn the user, but keep going.
2266 * Our failure will be noticed either in the next iteration of
2267 * the loop, or if this is the final delta, in the caller when
2268 * we return NULL. Those code paths will take care of making
2269 * a more explicit warning and retrying with another copy of
2273 error("failed to apply delta");
2281 unuse_pack(&w_curs
);
2285 const unsigned char *nth_packed_object_sha1(struct packed_git
*p
,
2288 const unsigned char *index
= p
->index_data
;
2290 if (open_pack_index(p
))
2292 index
= p
->index_data
;
2294 if (n
>= p
->num_objects
)
2297 if (p
->index_version
== 1) {
2298 return index
+ 24 * n
+ 4;
2301 return index
+ 20 * n
;
2305 off_t
nth_packed_object_offset(const struct packed_git
*p
, uint32_t n
)
2307 const unsigned char *index
= p
->index_data
;
2309 if (p
->index_version
== 1) {
2310 return ntohl(*((uint32_t *)(index
+ 24 * n
)));
2313 index
+= 8 + p
->num_objects
* (20 + 4);
2314 off
= ntohl(*((uint32_t *)(index
+ 4 * n
)));
2315 if (!(off
& 0x80000000))
2317 index
+= p
->num_objects
* 4 + (off
& 0x7fffffff) * 8;
2318 return (((uint64_t)ntohl(*((uint32_t *)(index
+ 0)))) << 32) |
2319 ntohl(*((uint32_t *)(index
+ 4)));
2323 off_t
find_pack_entry_one(const unsigned char *sha1
,
2324 struct packed_git
*p
)
2326 const uint32_t *level1_ofs
= p
->index_data
;
2327 const unsigned char *index
= p
->index_data
;
2328 unsigned hi
, lo
, stride
;
2329 static int use_lookup
= -1;
2330 static int debug_lookup
= -1;
2332 if (debug_lookup
< 0)
2333 debug_lookup
= !!getenv("GIT_DEBUG_LOOKUP");
2336 if (open_pack_index(p
))
2338 level1_ofs
= p
->index_data
;
2339 index
= p
->index_data
;
2341 if (p
->index_version
> 1) {
2346 hi
= ntohl(level1_ofs
[*sha1
]);
2347 lo
= ((*sha1
== 0x0) ? 0 : ntohl(level1_ofs
[*sha1
- 1]));
2348 if (p
->index_version
> 1) {
2356 printf("%02x%02x%02x... lo %u hi %u nr %"PRIu32
"\n",
2357 sha1
[0], sha1
[1], sha1
[2], lo
, hi
, p
->num_objects
);
2360 use_lookup
= !!getenv("GIT_USE_LOOKUP");
2362 int pos
= sha1_entry_pos(index
, stride
, 0,
2363 lo
, hi
, p
->num_objects
, sha1
);
2366 return nth_packed_object_offset(p
, pos
);
2370 unsigned mi
= (lo
+ hi
) / 2;
2371 int cmp
= hashcmp(index
+ mi
* stride
, sha1
);
2374 printf("lo %u hi %u rg %u mi %u\n",
2375 lo
, hi
, hi
- lo
, mi
);
2377 return nth_packed_object_offset(p
, mi
);
2386 int is_pack_valid(struct packed_git
*p
)
2388 /* An already open pack is known to be valid. */
2389 if (p
->pack_fd
!= -1)
2392 /* If the pack has one window completely covering the
2393 * file size, the pack is known to be valid even if
2394 * the descriptor is not currently open.
2397 struct pack_window
*w
= p
->windows
;
2399 if (!w
->offset
&& w
->len
== p
->pack_size
)
2403 /* Force the pack to open to prove its valid. */
2404 return !open_packed_git(p
);
2407 static int fill_pack_entry(const unsigned char *sha1
,
2408 struct pack_entry
*e
,
2409 struct packed_git
*p
)
2413 if (p
->num_bad_objects
) {
2415 for (i
= 0; i
< p
->num_bad_objects
; i
++)
2416 if (!hashcmp(sha1
, p
->bad_object_sha1
+ 20 * i
))
2420 offset
= find_pack_entry_one(sha1
, p
);
2425 * We are about to tell the caller where they can locate the
2426 * requested object. We better make sure the packfile is
2427 * still here and can be accessed before supplying that
2428 * answer, as it may have been deleted since the index was
2431 if (!is_pack_valid(p
)) {
2432 warning("packfile %s cannot be accessed", p
->pack_name
);
2437 hashcpy(e
->sha1
, sha1
);
2441 static int find_pack_entry(const unsigned char *sha1
, struct pack_entry
*e
)
2443 struct packed_git
*p
;
2445 prepare_packed_git();
2449 if (last_found_pack
&& fill_pack_entry(sha1
, e
, last_found_pack
))
2452 for (p
= packed_git
; p
; p
= p
->next
) {
2453 if (p
== last_found_pack
|| !fill_pack_entry(sha1
, e
, p
))
2456 last_found_pack
= p
;
2462 struct packed_git
*find_sha1_pack(const unsigned char *sha1
,
2463 struct packed_git
*packs
)
2465 struct packed_git
*p
;
2467 for (p
= packs
; p
; p
= p
->next
) {
2468 if (find_pack_entry_one(sha1
, p
))
2475 static int sha1_loose_object_info(const unsigned char *sha1
,
2476 struct object_info
*oi
)
2479 unsigned long mapsize
, size
;
2485 * If we don't care about type or size, then we don't
2486 * need to look inside the object at all.
2488 if (!oi
->typep
&& !oi
->sizep
) {
2489 if (oi
->disk_sizep
) {
2491 if (stat_sha1_file(sha1
, &st
) < 0)
2493 *oi
->disk_sizep
= st
.st_size
;
2498 map
= map_sha1_file(sha1
, &mapsize
);
2502 *oi
->disk_sizep
= mapsize
;
2503 if (unpack_sha1_header(&stream
, map
, mapsize
, hdr
, sizeof(hdr
)) < 0)
2504 status
= error("unable to unpack %s header",
2506 else if ((status
= parse_sha1_header(hdr
, &size
)) < 0)
2507 status
= error("unable to parse %s header", sha1_to_hex(sha1
));
2510 git_inflate_end(&stream
);
2511 munmap(map
, mapsize
);
2513 *oi
->typep
= status
;
2517 int sha1_object_info_extended(const unsigned char *sha1
, struct object_info
*oi
)
2519 struct cached_object
*co
;
2520 struct pack_entry e
;
2523 co
= find_cached_object(sha1
);
2526 *(oi
->typep
) = co
->type
;
2528 *(oi
->sizep
) = co
->size
;
2530 *(oi
->disk_sizep
) = 0;
2531 oi
->whence
= OI_CACHED
;
2535 if (!find_pack_entry(sha1
, &e
)) {
2536 /* Most likely it's a loose object. */
2537 if (!sha1_loose_object_info(sha1
, oi
)) {
2538 oi
->whence
= OI_LOOSE
;
2542 /* Not a loose object; someone else may have just packed it. */
2543 reprepare_packed_git();
2544 if (!find_pack_entry(sha1
, &e
))
2548 rtype
= packed_object_info(e
.p
, e
.offset
, oi
);
2550 mark_bad_packed_object(e
.p
, sha1
);
2551 return sha1_object_info_extended(sha1
, oi
);
2552 } else if (in_delta_base_cache(e
.p
, e
.offset
)) {
2553 oi
->whence
= OI_DBCACHED
;
2555 oi
->whence
= OI_PACKED
;
2556 oi
->u
.packed
.offset
= e
.offset
;
2557 oi
->u
.packed
.pack
= e
.p
;
2558 oi
->u
.packed
.is_delta
= (rtype
== OBJ_REF_DELTA
||
2559 rtype
== OBJ_OFS_DELTA
);
2565 /* returns enum object_type or negative */
2566 int sha1_object_info(const unsigned char *sha1
, unsigned long *sizep
)
2568 enum object_type type
;
2569 struct object_info oi
= {NULL
};
2573 if (sha1_object_info_extended(sha1
, &oi
) < 0)
2578 static void *read_packed_sha1(const unsigned char *sha1
,
2579 enum object_type
*type
, unsigned long *size
)
2581 struct pack_entry e
;
2584 if (!find_pack_entry(sha1
, &e
))
2586 data
= cache_or_unpack_entry(e
.p
, e
.offset
, size
, type
, 1);
2589 * We're probably in deep shit, but let's try to fetch
2590 * the required object anyway from another pack or loose.
2591 * This should happen only in the presence of a corrupted
2592 * pack, and is better than failing outright.
2594 error("failed to read object %s at offset %"PRIuMAX
" from %s",
2595 sha1_to_hex(sha1
), (uintmax_t)e
.offset
, e
.p
->pack_name
);
2596 mark_bad_packed_object(e
.p
, sha1
);
2597 data
= read_object(sha1
, type
, size
);
2602 int pretend_sha1_file(void *buf
, unsigned long len
, enum object_type type
,
2603 unsigned char *sha1
)
2605 struct cached_object
*co
;
2607 hash_sha1_file(buf
, len
, typename(type
), sha1
);
2608 if (has_sha1_file(sha1
) || find_cached_object(sha1
))
2610 if (cached_object_alloc
<= cached_object_nr
) {
2611 cached_object_alloc
= alloc_nr(cached_object_alloc
);
2612 cached_objects
= xrealloc(cached_objects
,
2613 sizeof(*cached_objects
) *
2614 cached_object_alloc
);
2616 co
= &cached_objects
[cached_object_nr
++];
2619 co
->buf
= xmalloc(len
);
2620 memcpy(co
->buf
, buf
, len
);
2621 hashcpy(co
->sha1
, sha1
);
2625 static void *read_object(const unsigned char *sha1
, enum object_type
*type
,
2626 unsigned long *size
)
2628 unsigned long mapsize
;
2630 struct cached_object
*co
;
2632 co
= find_cached_object(sha1
);
2636 return xmemdupz(co
->buf
, co
->size
);
2639 buf
= read_packed_sha1(sha1
, type
, size
);
2642 map
= map_sha1_file(sha1
, &mapsize
);
2644 buf
= unpack_sha1_file(map
, mapsize
, type
, size
, sha1
);
2645 munmap(map
, mapsize
);
2648 reprepare_packed_git();
2649 return read_packed_sha1(sha1
, type
, size
);
2653 * This function dies on corrupt objects; the callers who want to
2654 * deal with them should arrange to call read_object() and give error
2655 * messages themselves.
2657 void *read_sha1_file_extended(const unsigned char *sha1
,
2658 enum object_type
*type
,
2659 unsigned long *size
,
2664 const struct packed_git
*p
;
2665 const unsigned char *repl
= lookup_replace_object_extended(sha1
, flag
);
2668 data
= read_object(repl
, type
, size
);
2672 if (errno
&& errno
!= ENOENT
)
2673 die_errno("failed to read object %s", sha1_to_hex(sha1
));
2675 /* die if we replaced an object with one that does not exist */
2677 die("replacement %s not found for %s",
2678 sha1_to_hex(repl
), sha1_to_hex(sha1
));
2680 if (has_loose_object(repl
)) {
2681 path
= sha1_file_name(sha1
);
2682 die("loose object %s (stored in %s) is corrupt",
2683 sha1_to_hex(repl
), path
);
2686 if ((p
= has_packed_and_bad(repl
)) != NULL
)
2687 die("packed object %s (stored in %s) is corrupt",
2688 sha1_to_hex(repl
), p
->pack_name
);
2693 void *read_object_with_reference(const unsigned char *sha1
,
2694 const char *required_type_name
,
2695 unsigned long *size
,
2696 unsigned char *actual_sha1_return
)
2698 enum object_type type
, required_type
;
2700 unsigned long isize
;
2701 unsigned char actual_sha1
[20];
2703 required_type
= type_from_string(required_type_name
);
2704 hashcpy(actual_sha1
, sha1
);
2706 int ref_length
= -1;
2707 const char *ref_type
= NULL
;
2709 buffer
= read_sha1_file(actual_sha1
, &type
, &isize
);
2712 if (type
== required_type
) {
2714 if (actual_sha1_return
)
2715 hashcpy(actual_sha1_return
, actual_sha1
);
2718 /* Handle references */
2719 else if (type
== OBJ_COMMIT
)
2721 else if (type
== OBJ_TAG
)
2722 ref_type
= "object ";
2727 ref_length
= strlen(ref_type
);
2729 if (ref_length
+ 40 > isize
||
2730 memcmp(buffer
, ref_type
, ref_length
) ||
2731 get_sha1_hex((char *) buffer
+ ref_length
, actual_sha1
)) {
2736 /* Now we have the ID of the referred-to object in
2737 * actual_sha1. Check again. */
2741 static void write_sha1_file_prepare(const void *buf
, unsigned long len
,
2742 const char *type
, unsigned char *sha1
,
2743 char *hdr
, int *hdrlen
)
2747 /* Generate the header */
2748 *hdrlen
= sprintf(hdr
, "%s %lu", type
, len
)+1;
2752 git_SHA1_Update(&c
, hdr
, *hdrlen
);
2753 git_SHA1_Update(&c
, buf
, len
);
2754 git_SHA1_Final(sha1
, &c
);
2758 * Move the just written object into its final resting place.
2759 * NEEDSWORK: this should be renamed to finalize_temp_file() as
2760 * "moving" is only a part of what it does, when no patch between
2761 * master to pu changes the call sites of this function.
2763 int move_temp_to_file(const char *tmpfile
, const char *filename
)
2767 if (object_creation_mode
== OBJECT_CREATION_USES_RENAMES
)
2769 else if (link(tmpfile
, filename
))
2773 * Coda hack - coda doesn't like cross-directory links,
2774 * so we fall back to a rename, which will mean that it
2775 * won't be able to check collisions, but that's not a
2778 * The same holds for FAT formatted media.
2780 * When this succeeds, we just return. We have nothing
2783 if (ret
&& ret
!= EEXIST
) {
2785 if (!rename(tmpfile
, filename
))
2789 unlink_or_warn(tmpfile
);
2791 if (ret
!= EEXIST
) {
2792 return error("unable to write sha1 filename %s: %s", filename
, strerror(ret
));
2794 /* FIXME!!! Collision check here ? */
2798 if (adjust_shared_perm(filename
))
2799 return error("unable to set permission to '%s'", filename
);
2803 static int write_buffer(int fd
, const void *buf
, size_t len
)
2805 if (write_in_full(fd
, buf
, len
) < 0)
2806 return error("file write error (%s)", strerror(errno
));
2810 int hash_sha1_file(const void *buf
, unsigned long len
, const char *type
,
2811 unsigned char *sha1
)
2815 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2819 /* Finalize a file on disk, and close it. */
2820 static void close_sha1_file(int fd
)
2822 if (fsync_object_files
)
2823 fsync_or_die(fd
, "sha1 file");
2825 die_errno("error when closing sha1 file");
2828 /* Size of directory component, including the ending '/' */
2829 static inline int directory_size(const char *filename
)
2831 const char *s
= strrchr(filename
, '/');
2834 return s
- filename
+ 1;
2838 * This creates a temporary file in the same directory as the final
2841 * We want to avoid cross-directory filename renames, because those
2842 * can have problems on various filesystems (FAT, NFS, Coda).
2844 static int create_tmpfile(char *buffer
, size_t bufsiz
, const char *filename
)
2846 int fd
, dirlen
= directory_size(filename
);
2848 if (dirlen
+ 20 > bufsiz
) {
2849 errno
= ENAMETOOLONG
;
2852 memcpy(buffer
, filename
, dirlen
);
2853 strcpy(buffer
+ dirlen
, "tmp_obj_XXXXXX");
2854 fd
= git_mkstemp_mode(buffer
, 0444);
2855 if (fd
< 0 && dirlen
&& errno
== ENOENT
) {
2856 /* Make sure the directory exists */
2857 memcpy(buffer
, filename
, dirlen
);
2858 buffer
[dirlen
-1] = 0;
2859 if (mkdir(buffer
, 0777) || adjust_shared_perm(buffer
))
2863 strcpy(buffer
+ dirlen
- 1, "/tmp_obj_XXXXXX");
2864 fd
= git_mkstemp_mode(buffer
, 0444);
2869 static int write_loose_object(const unsigned char *sha1
, char *hdr
, int hdrlen
,
2870 const void *buf
, unsigned long len
, time_t mtime
)
2873 unsigned char compressed
[4096];
2876 unsigned char parano_sha1
[20];
2878 static char tmp_file
[PATH_MAX
];
2880 filename
= sha1_file_name(sha1
);
2881 fd
= create_tmpfile(tmp_file
, sizeof(tmp_file
), filename
);
2883 if (errno
== EACCES
)
2884 return error("insufficient permission for adding an object to repository database %s", get_object_directory());
2886 return error("unable to create temporary file: %s", strerror(errno
));
2890 memset(&stream
, 0, sizeof(stream
));
2891 git_deflate_init(&stream
, zlib_compression_level
);
2892 stream
.next_out
= compressed
;
2893 stream
.avail_out
= sizeof(compressed
);
2896 /* First header.. */
2897 stream
.next_in
= (unsigned char *)hdr
;
2898 stream
.avail_in
= hdrlen
;
2899 while (git_deflate(&stream
, 0) == Z_OK
)
2901 git_SHA1_Update(&c
, hdr
, hdrlen
);
2903 /* Then the data itself.. */
2904 stream
.next_in
= (void *)buf
;
2905 stream
.avail_in
= len
;
2907 unsigned char *in0
= stream
.next_in
;
2908 ret
= git_deflate(&stream
, Z_FINISH
);
2909 git_SHA1_Update(&c
, in0
, stream
.next_in
- in0
);
2910 if (write_buffer(fd
, compressed
, stream
.next_out
- compressed
) < 0)
2911 die("unable to write sha1 file");
2912 stream
.next_out
= compressed
;
2913 stream
.avail_out
= sizeof(compressed
);
2914 } while (ret
== Z_OK
);
2916 if (ret
!= Z_STREAM_END
)
2917 die("unable to deflate new object %s (%d)", sha1_to_hex(sha1
), ret
);
2918 ret
= git_deflate_end_gently(&stream
);
2920 die("deflateEnd on object %s failed (%d)", sha1_to_hex(sha1
), ret
);
2921 git_SHA1_Final(parano_sha1
, &c
);
2922 if (hashcmp(sha1
, parano_sha1
) != 0)
2923 die("confused by unstable object source data for %s", sha1_to_hex(sha1
));
2925 close_sha1_file(fd
);
2930 utb
.modtime
= mtime
;
2931 if (utime(tmp_file
, &utb
) < 0)
2932 warning("failed utime() on %s: %s",
2933 tmp_file
, strerror(errno
));
2936 return move_temp_to_file(tmp_file
, filename
);
2939 int write_sha1_file(const void *buf
, unsigned long len
, const char *type
, unsigned char *returnsha1
)
2941 unsigned char sha1
[20];
2945 /* Normally if we have it in the pack then we do not bother writing
2946 * it out into .git/objects/??/?{38} file.
2948 write_sha1_file_prepare(buf
, len
, type
, sha1
, hdr
, &hdrlen
);
2950 hashcpy(returnsha1
, sha1
);
2951 if (has_sha1_file(sha1
))
2953 return write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, 0);
2956 int force_object_loose(const unsigned char *sha1
, time_t mtime
)
2960 enum object_type type
;
2965 if (has_loose_object(sha1
))
2967 buf
= read_packed_sha1(sha1
, &type
, &len
);
2969 return error("cannot read sha1_file for %s", sha1_to_hex(sha1
));
2970 hdrlen
= sprintf(hdr
, "%s %lu", typename(type
), len
) + 1;
2971 ret
= write_loose_object(sha1
, hdr
, hdrlen
, buf
, len
, mtime
);
2977 int has_pack_index(const unsigned char *sha1
)
2980 if (stat(sha1_pack_index_name(sha1
), &st
))
2985 int has_sha1_pack(const unsigned char *sha1
)
2987 struct pack_entry e
;
2988 return find_pack_entry(sha1
, &e
);
2991 int has_sha1_file(const unsigned char *sha1
)
2993 struct pack_entry e
;
2995 if (find_pack_entry(sha1
, &e
))
2997 if (has_loose_object(sha1
))
2999 reprepare_packed_git();
3000 return find_pack_entry(sha1
, &e
);
3003 static void check_tree(const void *buf
, size_t size
)
3005 struct tree_desc desc
;
3006 struct name_entry entry
;
3008 init_tree_desc(&desc
, buf
, size
);
3009 while (tree_entry(&desc
, &entry
))
3011 * tree_entry() will die() on malformed entries */
3015 static void check_commit(const void *buf
, size_t size
)
3018 memset(&c
, 0, sizeof(c
));
3019 if (parse_commit_buffer(&c
, buf
, size
))
3020 die("corrupt commit");
3023 static void check_tag(const void *buf
, size_t size
)
3026 memset(&t
, 0, sizeof(t
));
3027 if (parse_tag_buffer(&t
, buf
, size
))
3031 static int index_mem(unsigned char *sha1
, void *buf
, size_t size
,
3032 enum object_type type
,
3033 const char *path
, unsigned flags
)
3035 int ret
, re_allocated
= 0;
3036 int write_object
= flags
& HASH_WRITE_OBJECT
;
3042 * Convert blobs to git internal format
3044 if ((type
== OBJ_BLOB
) && path
) {
3045 struct strbuf nbuf
= STRBUF_INIT
;
3046 if (convert_to_git(path
, buf
, size
, &nbuf
,
3047 write_object
? safe_crlf
: SAFE_CRLF_FALSE
)) {
3048 buf
= strbuf_detach(&nbuf
, &size
);
3052 if (flags
& HASH_FORMAT_CHECK
) {
3053 if (type
== OBJ_TREE
)
3054 check_tree(buf
, size
);
3055 if (type
== OBJ_COMMIT
)
3056 check_commit(buf
, size
);
3057 if (type
== OBJ_TAG
)
3058 check_tag(buf
, size
);
3062 ret
= write_sha1_file(buf
, size
, typename(type
), sha1
);
3064 ret
= hash_sha1_file(buf
, size
, typename(type
), sha1
);
3070 static int index_pipe(unsigned char *sha1
, int fd
, enum object_type type
,
3071 const char *path
, unsigned flags
)
3073 struct strbuf sbuf
= STRBUF_INIT
;
3076 if (strbuf_read(&sbuf
, fd
, 4096) >= 0)
3077 ret
= index_mem(sha1
, sbuf
.buf
, sbuf
.len
, type
, path
, flags
);
3080 strbuf_release(&sbuf
);
3084 #define SMALL_FILE_SIZE (32*1024)
3086 static int index_core(unsigned char *sha1
, int fd
, size_t size
,
3087 enum object_type type
, const char *path
,
3093 ret
= index_mem(sha1
, NULL
, size
, type
, path
, flags
);
3094 } else if (size
<= SMALL_FILE_SIZE
) {
3095 char *buf
= xmalloc(size
);
3096 if (size
== read_in_full(fd
, buf
, size
))
3097 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3099 ret
= error("short read %s", strerror(errno
));
3102 void *buf
= xmmap(NULL
, size
, PROT_READ
, MAP_PRIVATE
, fd
, 0);
3103 ret
= index_mem(sha1
, buf
, size
, type
, path
, flags
);
3110 * This creates one packfile per large blob unless bulk-checkin
3111 * machinery is "plugged".
3113 * This also bypasses the usual "convert-to-git" dance, and that is on
3114 * purpose. We could write a streaming version of the converting
3115 * functions and insert that before feeding the data to fast-import
3116 * (or equivalent in-core API described above). However, that is
3117 * somewhat complicated, as we do not know the size of the filter
3118 * result, which we need to know beforehand when writing a git object.
3119 * Since the primary motivation for trying to stream from the working
3120 * tree file and to avoid mmaping it in core is to deal with large
3121 * binary blobs, they generally do not want to get any conversion, and
3122 * callers should avoid this code path when filters are requested.
3124 static int index_stream(unsigned char *sha1
, int fd
, size_t size
,
3125 enum object_type type
, const char *path
,
3128 return index_bulk_checkin(sha1
, fd
, size
, type
, path
, flags
);
3131 int index_fd(unsigned char *sha1
, int fd
, struct stat
*st
,
3132 enum object_type type
, const char *path
, unsigned flags
)
3135 size_t size
= xsize_t(st
->st_size
);
3137 if (!S_ISREG(st
->st_mode
))
3138 ret
= index_pipe(sha1
, fd
, type
, path
, flags
);
3139 else if (size
<= big_file_threshold
|| type
!= OBJ_BLOB
||
3140 (path
&& would_convert_to_git(path
, NULL
, 0, 0)))
3141 ret
= index_core(sha1
, fd
, size
, type
, path
, flags
);
3143 ret
= index_stream(sha1
, fd
, size
, type
, path
, flags
);
3148 int index_path(unsigned char *sha1
, const char *path
, struct stat
*st
, unsigned flags
)
3151 struct strbuf sb
= STRBUF_INIT
;
3153 switch (st
->st_mode
& S_IFMT
) {
3155 fd
= open(path
, O_RDONLY
);
3157 return error("open(\"%s\"): %s", path
,
3159 if (index_fd(sha1
, fd
, st
, OBJ_BLOB
, path
, flags
) < 0)
3160 return error("%s: failed to insert into database",
3164 if (strbuf_readlink(&sb
, path
, st
->st_size
)) {
3165 char *errstr
= strerror(errno
);
3166 return error("readlink(\"%s\"): %s", path
,
3169 if (!(flags
& HASH_WRITE_OBJECT
))
3170 hash_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
);
3171 else if (write_sha1_file(sb
.buf
, sb
.len
, blob_type
, sha1
))
3172 return error("%s: failed to insert into database",
3174 strbuf_release(&sb
);
3177 return resolve_gitlink_ref(path
, "HEAD", sha1
);
3179 return error("%s: unsupported file type", path
);
3184 int read_pack_header(int fd
, struct pack_header
*header
)
3186 if (read_in_full(fd
, header
, sizeof(*header
)) < sizeof(*header
))
3187 /* "eof before pack header was fully read" */
3188 return PH_ERROR_EOF
;
3190 if (header
->hdr_signature
!= htonl(PACK_SIGNATURE
))
3191 /* "protocol error (pack signature mismatch detected)" */
3192 return PH_ERROR_PACK_SIGNATURE
;
3193 if (!pack_version_ok(header
->hdr_version
))
3194 /* "protocol error (pack version unsupported)" */
3195 return PH_ERROR_PROTOCOL
;
3199 void assert_sha1_type(const unsigned char *sha1
, enum object_type expect
)
3201 enum object_type type
= sha1_object_info(sha1
, NULL
);
3203 die("%s is not a valid object", sha1_to_hex(sha1
));
3205 die("%s is not a valid '%s' object", sha1_to_hex(sha1
),